OpenAI and Anthropic are facing criticism over AI and copyright from Australia’s music industry, with a Law Commentary headline describing warnings of AI exploitation. The available material does not identify the music organisations involved, set out their allegations or provide responses from either company.
The report’s headline points to a dispute over AI training and copyright, but the accompanying excerpt concerns an unrelated lawsuit involving OpenAI and cybersecurity testing. The three supplied entries are identical and link to the same page, rather than providing independent accounts. That mismatch leaves the specific Australian claims, any proposed legal changes and the companies’ positions unverified.
The broader legal issue is how copyright applies when developers use protected material to build AI systems. Training typically involves processing large collections of data to identify statistical patterns. Depending on the system, that material can include text, images, audio or other content. Whether particular uses require permission depends on the applicable law, the rights involved and the way the material is obtained and processed.
Music presents several distinct rights questions. A song’s composition and lyrics can be protected separately from a recording of its performance. Publishers, songwriters, performers and recording owners may therefore have different interests in the same piece of music. Permission covering one layer does not necessarily resolve questions about another, making the scope of any training licence important.
Australian copyright law provides specific exceptions for certain uses of protected works, rather than the broad, open-ended fair-use framework found in the United States. That distinction matters when assessing arguments about AI training: a legal position advanced in one country cannot simply be assumed to apply in another. The relevant analysis also depends on which acts occurred and where they took place.
Training and generated output raise related but separate questions. A dispute about copying material into a training dataset is not identical to a claim that a system’s output reproduces protected expression. Similarity in genre or style alone does not establish that a particular copyrighted work has been infringed. Evaluating an output requires attention to the actual material reproduced and the relevant legal standards.
Licensing is one possible route for addressing the use of copyrighted music in AI development. Such arrangements can define permitted uses, payment terms and obligations concerning the material supplied. Transparency about training data can also affect rights holders’ ability to establish whether their works were used, although transparency alone does not settle whether a use was lawful.
For this reported Australian backlash, the central details remain missing: who raised the objections, which uses they challenged, whether they sought licensing payments or legislative action, and how OpenAI and Anthropic responded. The supplied material does not establish that an Australian lawsuit has been filed or that a court has ruled on the alleged conduct.
Sources: Law Commentary